When I first examined the cheap XH-W1401 thermostat module (6 digit 7S-LED display, 4 keys, 2 LEDs, relay, and buzzer) I quickly noticed that using the GPIO PD1/SWIM for 3 tasks (SWIM for flashing, half-duplex RX and TX) wouldn't be that easy since it already serves for SWIM and to clock the 74HC164 shift register.
Now I found a solution: the shift register clock pulse can be short with respect to a UART bit time. I tested it with a two cycle operation (125ns) which a CH340 UART chip filtered out. First tests look promising!
It's also possible to generate the clock with one cycle (64ns) and to use a very simple RC filter (e.g. 47R/10nF) to hide it even from an unfiltered UART like the interrupt driven implementation in this project.
It's further complicated by the fact that the 74HC164 is unlatched which means that any transition on the serial line alters the LED pattern of a currently displayed digit. This can be countered by refreshing the display contents in the stop bit (sometimes even a software UART has an advantage ;-) ). The GPIO will then serve 4 needs concurrently: quadruple play ;-)
It looks like the XH-W1401 has the potential to be a supported gadget (there are variants with a front panel, and at a price of about $2.60 it's still very cheap). The schematics still has some mysteries (e.g. what they did at the sensor input, and which STM8s pins are assigned to keys, relay and buzzer) but if you follow this little project ordering a W1401 now wouldn't be all too risky.
Edit1: I tested optimized assembly code for feeding the 74HC164 with 62ns pulses, and I was mildly surprised by the bit rate. Depending on instruction alignment with 32bit boundaries I get 1.75 to 2 Mbit/sec with looped code (data shift out, clock creation, counting, branching). In my youth that would have been ... fast ;-)
Edit2: the multiplexed LED display and the serial interface on the same port appear to work. For now, the display is off while a byte is being transmitted through Rx or Tx. One of the next things will be more examination of the circuit. Especially the NTC sensor input is unexpected, and a huge capacitor at NRST made me learn a new thing about ICP through SWIM: NRST isn't always neded!